Add extra accs argument for LogDensityFunction

[penelopeysm]

This PR fixes another feature regression in FastLDF, which is that the accumulators used for actual evaluation of the log-density are hardcoded. (Technically not hardcoded, but you have to overload some internal function, which is not good.)

In SlowLDF the accs were taken from the VarInfo passed in:

DynamicPPL.jl/src/logdensityfunction.jl

Lines 135 to 140 in 6c615ad

	function LogDensityFunction(
	model::Model,
	getlogdensity::Function=getlogjoint_internal,
	varinfo::AbstractVarInfo=ldf_default_varinfo(model, getlogdensity);
	adtype::Union{ADTypes.AbstractADType,Nothing}=nothing,
	)

I found that I needed this functionality in the optimisation code so that I can pass in the accumulator that checks at runtime whether constraints are satisfied.

I think this isn't breaking, since it's the final positional argument and it is optional.

[github-actions]

Benchmark Report

• this PR's head: 7e63b619e58d19b9a7005a188dd82373477085f4
• base branch: 47f6b7e9e72e23ba5f777d91b16c1f4064962992

Computer Information

Julia Version 1.11.8
Commit cf1da5e20e3 (2025-11-06 17:49 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

┌───────────────────────┬───────┬─────────────┬───────────────────┬────────┬─────────────────────────────────┬────────────────────────────┬─────────────────────────────────┐
│                       │       │             │                   │        │        t(eval) / t(ref)         │     t(grad) / t(eval)      │        t(grad) / t(ref)         │
│                       │       │             │                   │        │ ──────────┬───────────┬──────── │ ───────┬─────────┬──────── │ ──────────┬───────────┬──────── │
│                 Model │   Dim │  AD Backend │           VarInfo │ Linked │      base │   this PR │ speedup │   base │ this PR │ speedup │      base │   this PR │ speedup │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│               Dynamic │    10 │    mooncake │             typed │   true │    386.72 │    369.83 │    1.05 │  10.63 │   10.02 │    1.06 │   4109.95 │   3706.79 │    1.11 │
│                   LDA │    12 │ reversediff │             typed │   true │   2741.07 │   2630.28 │    1.04 │   4.94 │    5.01 │    0.98 │  13535.52 │  13188.03 │    1.03 │
│   Loop univariate 10k │ 10000 │    mooncake │             typed │   true │ 105343.13 │ 104025.91 │    1.01 │   4.14 │    3.83 │    1.08 │ 435985.15 │ 398535.16 │    1.09 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│    Loop univariate 1k │  1000 │    mooncake │             typed │   true │   7890.33 │   7961.12 │    0.99 │   4.76 │    4.59 │    1.04 │  37538.22 │  36544.55 │    1.03 │
│      Multivariate 10k │ 10000 │    mooncake │             typed │   true │  30832.29 │  33311.59 │    0.93 │  10.55 │    9.89 │    1.07 │ 325340.55 │ 329366.03 │    0.99 │
│       Multivariate 1k │  1000 │    mooncake │             typed │   true │   3630.62 │   3620.74 │    1.00 │  12.88 │    9.20 │    1.40 │  46760.18 │  33304.56 │    1.40 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│ Simple assume observe │     1 │ forwarddiff │             typed │  false │      2.73 │      2.66 │    1.03 │   3.92 │    3.98 │    0.99 │     10.72 │     10.60 │    1.01 │
│           Smorgasbord │   201 │ forwarddiff │             typed │  false │   1225.28 │   1210.56 │    1.01 │ 123.49 │  124.77 │    0.99 │ 151311.89 │ 151038.02 │    1.00 │
│           Smorgasbord │   201 │ forwarddiff │       simple_dict │   true │       err │       err │     err │    err │     err │     err │       err │       err │     err │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │ simple_namedtuple │   true │       err │       err │     err │    err │     err │     err │       err │       err │     err │
│           Smorgasbord │   201 │      enzyme │             typed │   true │   1674.72 │   1675.04 │    1.00 │   6.72 │    6.18 │    1.09 │  11255.59 │  10356.75 │    1.09 │
│           Smorgasbord │   201 │    mooncake │             typed │   true │   1672.35 │   2346.28 │    0.71 │   5.19 │    3.77 │    1.38 │   8676.14 │   8842.10 │    0.98 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ reversediff │             typed │   true │   1717.42 │   1662.47 │    1.03 │  89.21 │   93.20 │    0.96 │ 153218.69 │ 154935.49 │    0.99 │
│           Smorgasbord │   201 │ forwarddiff │      typed_vector │   true │   1673.14 │   1659.35 │    1.01 │  56.10 │   58.24 │    0.96 │  93865.83 │  96636.85 │    0.97 │
│           Smorgasbord │   201 │ forwarddiff │           untyped │   true │   1669.19 │   1654.66 │    1.01 │  58.12 │   58.30 │    1.00 │  97013.17 │  96463.03 │    1.01 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │    untyped_vector │   true │   1683.50 │   1674.34 │    1.01 │  55.73 │   57.53 │    0.97 │  93827.79 │  96316.85 │    0.97 │
│              Submodel │     1 │    mooncake │             typed │   true │      7.56 │      7.51 │    1.01 │   5.34 │    5.16 │    1.04 │     40.41 │     38.74 │    1.04 │
└───────────────────────┴───────┴─────────────┴───────────────────┴────────┴───────────┴───────────┴─────────┴────────┴─────────┴─────────┴───────────┴───────────┴─────────┘

[github-actions]

DynamicPPL.jl documentation for PR #1196 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1196/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.00%. Comparing base (47f6b7e) to head (7e63b61).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1196   +/-   ##
=======================================
  Coverage   78.99%   79.00%           
=======================================
  Files          41       41           
  Lines        3938     3939    +1     
=======================================
+ Hits         3111     3112    +1     
  Misses        827      827           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mhauru]

LGTM